I've been getting the Py_buffer protocol based format handler for PyOpenGL implemented. The actual handler for buffer-providing types is pretty simple, and the numpy handler is now rewritten into a tiny subclass that just provides a "zeros" method and some "coerce to contiguous" code.  Not yet ready for prime time, and needs accelerator support still, but it does seem to basically work.

I haven't looked at automatically populating the more complex array parameters from the strides/dims/etc when one is using an in memory (non-VBO) array... I may just punt on that because you shouldn't likely be doing that anyway (i.e. your data likely should be in a VBO anyway).